Effects of hydrometeorological and other factors on SARS-CoV-2 reproduction number in three contiguous countries of tropical Andean South America: a spatiotemporally disaggregated time series analysis

Evolving Drivers of Brazilian SARS‐CoV‐2 Transmission: A Spatiotemporally Disaggregated Time Series Analysis of Meteorology, Policy, and Human Mobility

“…Specifically, we fit our GAMs to daily, state-level R t assuming a Gaussian distribution with a log link. Our choice of a Gaussian family reflects the quasi-normal distribution of the outcome variable, R t (Figure S1 in Supporting Information S1) and is also consistent with recent work modeling R t with GAMs (Colston et al, 2023). For each time period of interest, our model has the form: (2) where t denotes time, which is day in our case; e is each Brazilian state ("estado" in Portuguese); 𝐴𝐴 lagged cases is the total number of confirmed COVID-19 cases during the preceding 30 days, which we include to account for autocorrelation; and 𝐴𝐴 𝐴𝐴(temperature) represents the standard deviation of temperature, used as a proxy for daily temperature variability.…”

“…Specifically, we fit our GAMs to daily, state‐level R t assuming a Gaussian distribution with a log link. Our choice of a Gaussian family reflects the quasi‐normal distribution of the outcome variable, R t (Figure S1 in Supporting Information ) and is also consistent with recent work modeling R t with GAMs (Colston et al., 2023). For each time period of interest, our model has the form: $$${R}_{t,e}\,\sim \,\mathrm{g}\mathrm{a}\mathrm{u}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{a}\mathrm{n}\left({\mu }_{t,e}\right)$$$ …”

“…We chose a relatively small number of model terms (Equation 2 ) compared with other studies which have included air pollutants (Wu et al., 2020 ); sociodemographic data and additional mobility indicators (Colston et al., 2023 ; Nottmeyer & Sera, 2021 ); and additional meteorological variables (Colston et al., 2023 ; Ma et al., 2021 ; Zhang et al., 2022 ). Our selected meteorological terms have an established precedent for shaping respiratory virus seasonality (Lowen & Steel, 2014 ) and were most‐commonly investigated in early studies on the meteorological drivers of SARS‐CoV‐2 transmission (Kerr et al., 2021 ).…”

“…Examining individual smooth functions allows us to see the impact of a single variable or interactions between variables on R t . GAMs have been extensively used to assess the environmental health outcomes and drivers of COVID‐19 variability (Colston et al., 2023 ; Dominici, 2002 ; Sera et al., 2021 ). Unlike other types of linear models that assume, for example, normality or no multicollinearity, GAMs are flexible and can handle non‐linear and non‐monotonic relationships between explanatory and response variables.…”

“…Seasonality and meteorology, particularly temperature and humidity, have been purported drivers of SARS‐CoV‐2 transmission based on their impact on aerosolization of virus droplets, virus survival on fomites, host susceptibility, and human behavior (Lowen & Steel, 2014 ; Tamerius et al., 2013 ; Yang et al., 2015 ). Yet, myriad early studies investigating the associations between meteorology and COVID‐19 have not always reached consistent findings regarding the role of meteorological factors (Colston et al., 2023 ; Ma et al., 2021 ; Sera et al., 2021 ), although these and other studies generally emphasize that while the associations between COVID‐19 and meteorological variables may be significant, they are small compared with disease control interventions and could not entirely explain excess disease burdens. Several factors have been suggested as reasons for these inconsistent findings: a short temporal data record; simplistic statistical frameworks such as correlation analyses that overlook confounding factors; and error‐prone variables such as case counts, which could be biased toward the null due to underreporting, testing delays, and the proliferation of at‐home testing (Kerr et al., 2021 ; Mecenas et al., 2020 ).…”